The aim of this study was to establish a data-driven nosology and to provide the neuroimaging correlates in FTLD. This analysis suggested a division into three groups, clearly distinguished by specific functional hypoperfusion patterns.
The clinical, pathological and genetic heterogeneity of FTLD accounts for its present nosological controversy, and opens the question of whether this generic label represents a unique syndrome best viewed as complex, or if it includes different and distinct entities.
At present, FTLD classification relies upon clinical symptoms at disease presentation, but the convergence of the different clinical pictures in overlapping syndromes and their poor concordance with pathology, made of the FTLD clinical diagnosis a challenging problem [9
]. We currently differentiate FTLD in bvFTD, SD, PNFA, PSP, CBD, and these clinical distinctions may also have functional correspondences (see Figure ). The mismatch, however, between clinical pictures and pathology makes the present classification largely unsatisfactory. This suggests the need of an alternative explanation to describe FTDL discrete clinical presentations.
In this study, we tried to disentangle these issues, employing the open statistical approach provided by Latent Profile Analysis and applying it to a large sample of FTLD patients. Our aim was to single out which, and how many clinical presentations could be generated by a statistical analysis blinded to an extended cognitive/behavioural data set in FTLD patients.
Latent Profile Analysis posits that a heterogeneous group can be reduced to several homogeneous subgroups through evaluating and then minimizing the pairwise correlations among responses across multiple continuous variables. Thus, Latent Profile Analysis is capable of determining the number and composition of unobserved latent classes that produce observed data. This approach is particularly useful when there is evidence that certain symptoms co-aggregate at above normal levels (that is, symptoms that are beyond what is usually seen in patients who present certain syndrome patterns) to form distinct clusters.
In the present study, cognitive performance, the severity of behavioural symptoms, and functional impairment were examined in a large sample of FTLD patients, and three distinct clusters were detected, named "pseudomanic behaviour", "cognitive" and "pseudodepressed behaviour".
The first, i.e. LC1, was represented by greater behavioural disturbances, such as dishinibition and abnormal social conduct, as evidenced in FBI and NPI scores. The concomitant impairment in BADL and IADL in these patients is therefore the result of inadequacy or disinhibition rather than the consequence of the cognitive decline. The second LC2 profile was underscored by a "cognitive" endophenotype, mainly characterized by executive dysfunction. The third LC3 showed better performances in neuropsychological test scores compared to the other two LCs, and subtle behavioural abnormalities, mainly represented by depressive symptoms.
It is noteworthy to mention that motor impairment, measured by UPDRS-III scale, was not used to define the latent classes, and was uncorrelated with either cognitive or behavioural profiles, suggesting an independence between motor aspects and cognitive/behavioural performances.
Thus, referring to the nosology of FTLD, our data-driven approach identified three clinical syndromes. In the same vein, a recent experimental work applied a similar approach, i.e. Factor Analysis, to clarify the classification issue in Progressive Aphasia [2
As a second goal of the present study, the SPECT imaging modality and Statistical Parametric Mapping (SPM2) were used to identify functional correspondence of the FTLD as generated by Latent Profile Analysis. The SPM2 method was applied to verify if the generated LCs depended on different stage of the disease, or if they were the expression of three different endophenotypes.
The three clusters obtained by Latent Profile Analysis were underscored by specific functional correlates. Hypoperfusion in frontal medial and orbitobasal cortex was found in the "pseudomanic behaviour" endophenotype, subcortical brain region hypoperfusion was detected in the "cognitive" endophenotype, hypoperfusion in the dorsolateral frontal cortex and insula characterised the "pseudodepressed behaviour" endophenotype (Figure ). These data provide evidence that specific neurofunctional-symptom cluster relationship can be delineated in patients with FTLD, thus excluding that these latent classes represent different disease stages.
Interestingly, the pseudomanic and pseudodepressed endophenotypes paralleled previous behavioural and metabolic findings reported in a group of FTLD patients [34
]. The main subset of symptoms in "cognitive" endophenotye, such as executive dysfunctions, can well be attributed to impairment of the dorsolateral-frontal circuit at the basal ganglia level; in fact, when data were explored at a lower threshold (P < 0.005), not only subcortical brain regions, but also dorsolateral frontal cortex was involved (-32,58,16; T = 2.87, cluster size = 101).
An increasing number of studies evaluating brain tissue volume and metabolic function in dementia demonstrate that regional tissue loss, or hypometabolism, correlates with specific cognitive or behavioural impairment in FTLD [35
]. FTLD patients, however, had profuse cognitive and behavioural abnormalities, thus suggesting that all these symptoms could have had a common neurofunctional basis [36
]. To date, few studies in either dementia or focal lesions has systematically examined the functional correlates of clustered behavioural and neuropsychological symptoms, such as here conducted by Latent Profile Analysis, instead of individual symptoms. We also performed the SPM2 analysis on the basis of pre-defined criteria (see Figure ), confirming previous neuroimaging findings [7
The present results suggest that FTLD syndromes lie on a continuum rather then existing as independence entities. Latent Profile Analysis approach demonstrated that FTLD can be summarised into three different clinical categories with a poor concordance with the usual clinical classification. Indeed, despite of a higher level of concordance in diagnosis distribution between LC1 and bvFTD, the other two LCs, i.e. LC2 and LC3, are otherwise variably represented (see Figure ).
It has been recently reported that there is a convergence of the syndromes in the course of the disease [9
]. In the future, it would be interesting to evaluate the clinical and overlapping aspects of these clusters over time.
Notwithstanding, approaches based on standardised assessment, similar or different to those presented in this study, are crucial for a clear-cut description of the wide range of features, both behavioural and cognitive, that may characterise FTLD and related disorders.
The results of this study should be replicated in other sample size cross-sectional studies, and these should be performed to determine if the same endophenotypes remain stable across studies. We indeed acknowledge the need of prognostic outcomes to prove the usefulness of the present scheme. Moreover, neuropathological confirmation would be necessary to further understand the relationship between clinical and neuropathological endophenotypes. In an era of treatment that targets disease-mechanism, it would be desirable that Latent Profile Analysis will be used to further discriminate molecular/genetic determinants of FTLD pathology.